The tractability and performance of microsimulating human activity for transportation systems analysis

The activity-based approach to travel demand analysis recognizes that human activities dictate travel. Microsimulation of household activity patterns has gained significant attention as a method for modeling this activity participation. Existing approaches, however, focus on modeling how households solve the activity scheduling problem—how and when each household member should engage in particular activities to meet the needs of the household. This is a top-down approach that recognizes inherent causal links between members of a household but sacrifices modeling flexibility that is necessary for complex policy analysis.; This dissertation synthesizes dominant activity analysis theories with concepts from the social simulation and complex systems analysis literature to demonstrate that the motivation and constraints that shape activities are more directly embodied in the activity execution problem—how individuals interact with other entities in their environment to engage in activity. The scheduling problem is re-cast as the adaptive internal process that an individual uses to navigate through this interactive environment to achieve environmentally-derived payoffs.; Based on this theory, a microsimulation is described that focuses on the activity execution process. Such a bottom-up approach presents a problem of tractability. This dissertation solves this problem by describing activity execution using a model of negotiated interaction derived from the Contract Net Protocol for distributed computation. This model is shown to be tractable in terms of the number of negotiating individuals, given reasonable limitations on the negotiation process. Then, a complete agent-based model of an urban activity system is described based on this activity execution kernel. This general model is shown to be tractable in terms of the population size, given assumptions on how negotiations are initiated. Finally, results from experiments using candidate adaptive learning algorithms for agents operating in the microsimulation are presented to demonstrate the utility of the microsimulation approach.